AU2004218675B2

AU2004218675B2 - Protein scaffolds for antibody mimics and other binding proteins

Info

Publication number: AU2004218675B2
Application number: AU2004218675A
Authority: AU
Inventors: Dasa Lipovsek
Original assignee: Bristol Myers Squibb Co
Current assignee: Bristol Myers Squibb Co
Priority date: 1998-12-10
Filing date: 2004-10-08
Publication date: 2008-02-28
Anticipated expiration: 2019-12-09
Also published as: JP2002532072A; CY1110549T1; WO2000034784A9; EP1137941A4; ES2329959T3; CA2351346A1; JP4562286B2; EP1137941A1; CA2351346C; DK1137941T4; EP2154535A1; ES2329959T5; EP1137941B1; WO2000034784A1; DK1137941T3; AU2357200A; PT1137941E; AU775076B2; ATE439592T1; DE69941267D1

Abstract

Disclosed herein are proteins that include a fibronectin type III domain having at least one randomized loop. Also disclosed herein are nucleic acids encoding such proteins and the use of such proteins in methods for evolving novel compound-binding species and their ligands.

Description

Regulation 3.2

AUSTRALIA

Patents Act 1990 COMPLETE SPECIFICATION STANDARD PATENT

APPLICANT:

Invention Title: Compound Therapeutics, Inc.

PROTEIN SCAFFOLDS FOR ANTIBODY MIMICS AND OTHER BINDING PROTEINS The following statement is a full description of this invention, including the best method of performing it known to me: 0 This invention relates to protein scaffolds useful, for example, for the g eneration of products having novel binding characteristics.

o Proteins having relatively defined three-dimensional structures, commonly referred to as protein scaffolds, may be used as reagents for the design of engineered products. These scaffolds typically contain one or more regions which are amenable to specific or random sequence variation, and such sequence randomization is often carried out to produce libraries of proteins from which desired products may be selected. One particular area in which such scaffolds are useful is the field of antibody design.

A number of previous approaches to the manipulation of the mammalian immune system to obtain reagents or drugs have been attempted.

These have included injecting animals with antigens of interest to obtain mixtures of polyclonal antibodies reactive against specific antigens, production of monoclonal antibodies in bybridoma cell culture (Koehler and Milstein, Nature 256:495, 1975), modification of existing monoclonal antibodies to obtain new or optimized recognition properties, creation of novel antibody fragments with desirable binding characteristics, and randomization of single chain antibodies (created by connecting the variable regions of the heavy and light chains of antibody molecules with a flexible peptide linker) followed by selection for antigen binding by phage display (Clackson et al., Nature 352:624, 1991).

In addition, several non-immunoglobulin protein scaffolds have been proposed for obtaining proteins wit novel binding properties. For example, a 0 o -2ci 0 "minibody" scaffold, which is related to the immnunoglobulin fold, has been 0 designed by deleting three beta strands from a heavy chain variable domain of a 0o monoclonal antibody (Tramontano et al., J. Mol. Recognit. 7:9, 1994). This protein includes 61 residues and can be used to present two hypervariable 1" 5 loops. These two loops have been randomized and products selected for 0 00 antigen binding, but thus far the framework appears to have somewhat limited utility due to solubility problems. Another framework used to display loops has 0 o been tendamistat, a 74 residue, six-strand beta sheet sandwich held together by N two disulfide bonds (McConnell and Hoess, J. Mol. Biol. 250:460, 1995). This scaffold includes three loops, but, to date, only two of these loops have been examined for randomization potential.

Other proteins have been tested as frameworks and have been used to display randomized residues on alpha helical surfaces (Nord et al., Nat.

Biotechnol. 15:772, 1997; Nord et al., Protein Eng. 8:601, 1995), loops between alpha helices in alpha helix bundles (Ku and Schultz, Proc. Natl. Acad.

Sci. USA 92:6552, 1995), and loops constrained by disulfide bridges, such as those of the small protease inhibitors (Markland et at., Biochemistry 35:8045, 1996; Markland et al., Biochemistry 35:8058, 1996; Rottgen and Collins, Gene 164:243, 1995; Wang et al., J. Biol. Chem. 270:12250, 1995).

Summary of the Invention The present invention provides a new family of proteins capable of evolving to bind any compound of interest. These proteins, which make use of a fibronectin or fibronectin-like scaffold, function in a manner characteristic of natural or engineered antibodies (that is, polyclonal, monoclonal, or single-chain antibodies) and, in additidon, possess structural advantages.

Specifically, the structure of these antibody mimics has been designed for optimal folding, stability, and solubility, even under conditions which normally 0 o -3ci t lead to the loss of structure and function in antibodies.

These antibody mimics may be utilized for the purpose of designing 00 o proteins which are capable of binding to virtually any compound (for example, any protein) of interest. In particular, the fibronectin-based molecules 5 described herein may be used as scaffolds which are subjected to directed 00 evolution designed to randomize one or more of the three fibronectin loops which are analogous to the complementarity-determining regions (CDRs) of an oantibody variable region. Such a directed evolution approach results in the production of antibody-like molecules with high affinities for antigens of interest. In addition, the scaffolds described herein may be used to display defined exposed loops (for example, loops previously randomized and selected on the basis of antigen binding) in order to direct the evolution of molecules that bind to such introduced loops. A selection of this type may be carried out to identify recognition molecules for any individual CDR-like loop or, alternatively, for the recognition of two or all three CDR-like loops combined into a non-linear epitope.

Accordingly, the present invention features a protein that includes a fibronectin type III domain having at least one randomized loop, the protein being characterized by its ability to bind to a compound that is not bound by the corresponding naturally-occurring fibronectin.

In preferred embodiments, the fibronectin type Ill domain is a mammalian (for example, a human) fibronectin type III domain; and the protein includes the tenth module of the fibronectin type III 0 Fn3) domain. In such proteins, compound binding is preferably mediated by either one, two, or three 0 Fn3 loops. In other preferred embodiments, the second loop of 0 Fn3 may be extended in length relative to the naturally-occurring module, or the 0 Fn3 may lack an integrin-binding motif. In these molecules, the integrin-binding motif may be replaced by an amino acid sequence in which a basic amino acid-

I

O .4.

Sneutral amino acid-acidic amino acid sequence (in the N-terminal to C-terminal O direction) replaces the integrin-binding motif; one preferred sequence is serineo glycine-glutamate. In another preferred embodiment, the fibronectin type III domain-containing proteins of the invention lack disulfide bonds.

Any of the fibronectin typelldomain-containing proteins described O herein may be formulated as pan of a fusion protein (for example, a fusion Sprotein which further includes an immunoglobulin F, domain, a complement S protein, a toxin protein, or an albumin protein). In addition, any of the c.i fibronectin type III domain proteins may be covalently bound to a nucleic acid (for example, an RNA), and the nucleic acid may encode the protein.

Moreover, the protein may be a multimer, or, particularly if it lacks an integrinbinding motif, it may be formulated in a physiologically-acceptable carrier.

The present invention also includes features proteins that include a fibronectin type III domain having at least one mutation in a P-sheet sequence which changes the scaffold structure. Again, these proteins are characterized by their ability to bind to compound that are not bound by the corresponding naturally-occurring fibronectin.

In a related aspect, the invention further features nucleic acids encoding any of the proteins of the invention. In preferred embodiments, the nucleic acid is DNA or RNA.

In another related aspect, the invention also features a method for generating a protein which includes a fibronectin type III domain and which is pharmaceutically acceptable to a mammal, involving removing the integrinbinding domain of said fibronectin type III domain. This method may be applied to any of the fibronectin type III domain-containing proteins described above and is particularly useful for generating proteins for human therapeutic applications. The invention also features such fibronectin type III domaincontaining proteins which lack integrin-binding domains.

O In yet other related aspects, the invention features screening methods 00 which may be used to obtain or evolve randomized fibronectin type III proteins 0 capable of binding to compounds of interest, or to obtain or evolve compounds (for example, proteins) capable of binding to a particular protein containing a Va 5 randomized fibronectin type III motif. In addition, the invention features 00 screening procedures which combine these two methods, in any order, to obtain either compounds or proteins of interest.

o In particular, the first screening method, useful for the isolation or identification of randomized proteins of interest, involves contacting the compound with a candidate protein, the candidate protein including a fibronectin type III domain having at least one randomized loop, the contacting being carried out under conditions that allow compound-proteli complex formation; and obtaining, from the complex, the protein which binds to the compound.

The second screening method, for isolating or identifying a compound which binds to a protein having a randomized fibronectin type II domain, involves: contacting the protein with a candidate compound, the contacting being carried out under conditions that allow compound-protein complex fonnation; and obtaining, from the complex, the compound which binds to the protein.

In preferred embodiments, the methods further involve either randomizing at least one loop of the fibronectin type III domain of the protein obtained in step and repeating steps and using the further randomized protein, or modifying the compound obtained in step and repeating steps (a) and using the further modified compound. In addition, the compound is preferably a protein, and the fibronectin type I domain is preferably a mammalian (for example, a human) fibronectin type III domain. In other preferred embodiments, the protein includes the tenth module of the fibronectin 0 o -6c.i t type III domain and binding is mediated by one, two or three 0 °Fn3 O loops. In addition, the second loop of 'tFn3 may be extended in length relative 00 0 to the naturally-occurring module, or Fn3 may lack an integrin-binding motif.

Again, as described above, the integrin-binding motif may be replaced by an

V")

5 amino acid sequence in which a basic amino acid-neutral amino acid-acidic 00 amino acid sequence (in the N-termnninal to C-terminal direction) replaces the c' integrin-binding motif; one preferred sequence is serine-glycine-glutamate.

o The selection methods described herein may be carried out using any fibronectin type III domain-containing protein. For example, the fibronectin type III domain-containing protein may lack disulfide bonds, or may be formulated as part of a fusion protein (for example, a fusion protein which further includes an immunoglobulin F, domain, a complement protein, a toxin protein, or an albumin protein). In addition, selections may be carried out using the fibronectin type III domain proteins covalently bound to nucleic acids (for example, RNAs or any nucleic acid which encodes the protein). Moreover, the selections may be carried out using fibronectin domain-containing protein multimers.

Preferably, the selections involve the immobilization of the binding target on a solid support. Preferred solid supports include columns (for example, affinity columns, such as agarose columns) or microchips.

As used herein, by "fibronectin type III domain" is meant a domain having 7 or 8 beta strands which are distributed between two beta sheets, which themselves pack against each other to form the core of the protein, and further containing loops which connect the beta strands to each other and are solvent exposed. There are at least three such loops at each edge of the beta sheet sandwich, where the edge is the boundary of the protein perpendicular to the direction of the beta strands. Preferably, a fibronectin type III domain includes a sequence which exhibits at least 30% amino acid identity, and preferably at 0 o7 0 least 50% amino acid identity, to the sequence encoding the structure of the 0"Fn3 domain referred to as "lttg" (ID "lttg" (one ttg)) available from the 00 o Protein Data Base. Sequence identity referred to in this definition is determined by the Homology program, available from Molecular Simulation t- 5 (San Diego, CA). The invention further includes polymers of 0 Fn3-related molecules, which are an extension of the use of the monomer structure, whether or not the subunits of the polyprotein are identical or different in sequence.

oBy "naturally occurring fibronectin" is meant any fibronectin protein that is encoded by a living organism.

By "randomized" is meant including one or more amino acid alterations relative to a template sequence.

By a "protein" is meant any sequence of two or more amino acids, regardless of length, post-translation modification, or function. "Protein" and "peptide" are used interchangeably herein.

By "RNA" is meant a sequence of two or more covalently bonded, naturally occurring or modified ribonucleotides. One example of a modified RNA included within this term is phospborothioate

RNA.

By "DNA" is meant a sequence of two or more covalently bonded, naturally occurring or modified deoxyribonucleotides.

By a "nucleic acid" is meant any two or more covalently bonded nucleotides or nucleotide analogs or derivatives. As used herein, this term includes, without limitation, DNA, RNA, and PNA.

By "pharmaceutically acceptable" is meant a compound or protein that may be administered to an animal (for example, a marnmal) without significant adverse medical consequences.

By "physiologically acceptable carrier" is meant a carrier which does not have a significant detrimental impact on the treated host and which retains the therapeutic properties of the compound with which it is administered. One 1- 8 "o exemplary physiologically acceptable carrier is physiological saline. Other o physiologically acceptable carriers and their formulations are known to one 00 0 skilled in the art and are described, for example, in Remington's Pharmaceutical Sciences, (18' edition), ed. A. Gennaro, 1990, Mack Publishing 0 5 Company, Easton, PA, incorporated herein by reference.

IND

00 By "selecting" is meant substantially partitioning a molecule from Cl other molecules in a population. As used herein, a "selecting" step provides at Sleast a 2-fold, preferably, a 30-fold, more preferably, a 100-fold, and, most C preferably, a 1000-fold enrichment of a desired molecule relative to undesired molecules in a population following the selection step. A selection step may be repeated any number of times, and different types of selection steps may be combined in a given approach.

By "binding partner," as used herein, is meant any molecule which has a specific, covalent or non-covalent affinity for a portion of a desired compound (for example, protein) of interest. Examples of binding partners include, without limitation, members of antigen/antibody pairs, protein/inhibitor pairs, receptor/ligand pairs (for example cell surface receptor/ligand pairs, such as hormone receptor/peptide hormone pairs), enzyme/substrate pairs (for example, kinase/substrate pairs), lectin/carbohydrate pairs, oligomeric or heterooligomeric protein aggregates, DNA binding protein/DNA binding site pairs, RNA/protein pairs, and nucleic acid duplexes, heteroduplexes, or ligated strands, as well as any molecule which is capable of forming one or more covalent or non-covalent bonds (for example, disulfide bonds) with any portion of another molecule (for example, a compound or protein).

By a "solid support" is meant, without limitation, any column (or column material), bead, test tube, microtiter dish, solid particle (for example, agarose or sepharose), microchip (for example, silicon, silicon-glass, or gold chip), or membrane (for example, the membrane of a liposome or vesicle) to o. which an affinity complex may be bound, either directly or indirectly (for 00 oo example, through other binding partner intermediates such as other antibodies or Protein or in which an affinity complex may be embedded (for example, through a receptor or channel).

NO

00 c-i The present invention provides a number of advantages. For example, as described in more detail below, the present antibody mimics Ni exhibit improved biophysical properties, such as stability under reducing conditions and solubility at high concentrations. In addition, these molecules may be readily expressed and folded in prokaryotic systems, such as E. cali, in eukaryotic systems, such as yeast, and in in viro translation systems, such as the rabbit reticulocyte lysate system. Moreover, these molecules are extremely amenable to affinity maturation techniques involving multiple cycles of selection, including in 3ito selection using RNA-protein fusion technology (Roberts arndc Szostak, Proc. Nat]. Acad. Sci USA 94:12297, 1997; Szostak et al., US Patent No. 6,258,558 and US Patent No. 6,261,804; Szostak et al.

W098/31700), phage display (see, for example, Smith and Petrehko, Chem.

Rev. 97:317, 1997), and yeast display systems (see, for example, Boder and Wittrup, Nature Biotech. 15:553, 1997).

Other features and advantages of the present invention will be apparent from the following detailed description thereof, and from the claims.

Brief Description of the Drawings FIGURE 1 is a photograph showing a comparison between the structures of antibody heavy chain variable regions from camel, and llama-, in each of two orientations.

FIGURE 2 is a photograph showing a comparison between the

FU

O

ci O structures of the camel antibody heavy chain variable region, the llama antibody 00 heavy chain variable region, and a fibronectin type III module number 10 0 Fn3) (top).

FIGURE 3 is a photograph showing a fibronectin type III module 00 number 10 ('OFn3).

o 10 FIGURE 4 is a graph illustrating a sequence alignment between a ci fibronectin type III protein domain and related protein domains, FIGURE 5 is a photograph showing the structural similarities between a 10 Fn3 domain and 15 related proteins, including fibronectins, tenascins, collagens, and undulin.

FIGURE 6 is a photograph showing space filling models of fibronectin III modules 9 and 10, in each of two different orientations. The two modules and the integrin binding loop (RGB) are labeled.

FIGURE 7 is a photograph showing space filling models of fibronectin III modules 7-10, in each of three different orientations. The four modules are labeled.

FIGURE 8 is a photograph illustrating the formation, under different salt conditions, of RNA-protein fusions which include fibronectin type III domains.

FIGURE 9 is a series of photographs illustrating the selection of fibronectin type III domain-containing RNA-protein fusions, as measured by

I

i 0 o -11o PCR signal analysis.

C) FIGURE 10 is a graph illustrating an increase in the percent TNF-z 00 o binding during the selections described herein, as well as a comparison between RNA-protein fusion and free protein selections.

a 5 FIGURE 11 is a series of schematic representations showing IgG, 00 0 Fn3, Fn-CR 1

-CH

2

-CH

3 and Fn-CH 2

-CH

3 (clockwise from top left).

FIGURE 12 is a photograph showing a molecular model of Fn-CHuo CH 2

-CH

3 based on known three-dimensional structures of IgG (X-ray crystallography) and "Fn3 (NMR and X-ray crystallography).

Dtaled Decfin The novel antibody mimics described herein have been designed to be superior both to antibody-derived fragments and to non-antibody frameworks, for example, those frameworks described above.

The major advantage of these antibody mimics over antibody fragments is structural. These scaffolds are derived from whole, stable, and soluble structurai modules found in human body fluid proteins. Consequently, they exhibit better folding and thermostability properties than antibody fragments, whose creation involves the removal of parts of the antibody native fold, often exposing amino acid residues that, in an intact antibody, would be buried in a hydrophobic environment, such as an interface between variable and constant domains. Exposure of such hydrophobic residues to solvent increases the likelihood of aggregation.

In addition, the antibody mimics described herein have no disulfide bonds, which have been reported to retard or prevent proper folding of antibody fragments under certain conditions. Since the present scaffolds do not rely on disulfides for native fold stability, they are stable under reducing conditions, unlike antibodies and their fragments which unravel upon disulfide bond

I

o-12ci t breakdown.

C) Moreover, these fibronectin-based scaffolds provide the functional 00 o advantages of antibody molecules. In particular, despite the fact that the 'OFn3 module is not an immunoglobulin, its overall fold is close to that of the variable region of the IgG heavy chain (Figure making it possible to display the three 00 fibronectin loops analogous to CDRs in relative orientations similar to those of native antibodies. Because of this structure, the present antibody mimics o possess antigen binding properties that are similar in nature and affinity to those of antibodies, and a loop randomization and shuffling strategy may be employed in vitr that is similar to the process of affinity maturation of antibodies in nix.

There are now described below exemplary fibronectin-based scaffolds and their use for identifying, selecting, and evolving novel binding proteins as well as their target ligands. These examples are provided for the purpose of illustrating, and not limiting, the invention.

Fn3 Struclural Moif The antibody mimics of the present invention are based on the structure of a fibronectin module of type Ill (Fn3), a common domain found in mammalian blood and structural proteins. This domain occurs more than 400 times in the protein sequence database and has been estimated to occur in 2% of the proteins sequenced to date, including fibronectins, tenscin, intracellular cytoskeletal proteins, and prokaryotic enzymes (Bork and Doolittle, Proc. Natl.

Acad. Sci. USA 89:8990, 1992; Bork et al., Nature Biotech. 15:553, 1997; Meinke et al., J. Bacteriol. 175:1910, 1993; Watanabe et al., S. Biol. Chem.

265:15659, 1990). In particular, these scaffolds include, as templates, the tenth module of human Fn3 ("iFn3), which comprises 94 amino acid residues. The overall fold of this domain is closely related to that of the smallest functional o -13t antibody fragment, the variable region of the heavy chain, which comprises the entire antigen recognition unit in camel and llama IgG (Figure 1, The major 00 o differences between camel and llama domains and the '(Fn3 domain are that (i) 0 Fn3 has fewer beta strands (seven vs. nine) and (ii) the two beta sheets packed against each other are connected by a disulfide bridge in the camel and llama VaO 00 domains, but not in 0 Fn3.

The three loops of 'Fn3 corresponding to the antigen-binding loops o of the IgG heavy chain run between amino acid residues 21-31, 51-56, and 76-88 (Figure The length of the first and the third loop, I11 and 12 residues, respectively, fall within the range of the corresponding antigen-recognition loops found in antibody heavy chains, that is, 10-12 and 3-25 residues, respectively. Accordingly, once randomized and selected for high antigen affinity, these two loops make contacts with antigens equivalent to the contacts of the corresponding loops in antibodies.

In contrast, the second loop of 'Fn3 is only 6 residues long, whereas the corresponding loop in antibody heavy chains ranges from 16-19 residues.

To optimize antigen binding, therefore, the second loop of 0 Fn3 is preferably extended by 10-13 residues (in addition to being randomized) to obtain the greatest possible flexibility and affinity in antigen binding. Indeed, in general, the lengths as well as the sequences of the CDR-like loops of the antibody mimics may be randomized during in sutm or in yirn affinity maturation (as described in more detail below).

The tent bunion fibronectin type III domain, 0 Fn3, refolds rapidly even at low temperature; its backbone conformation has been recovered within 1 second at 5 0 C. Thermodynamic stability of 0 Fn3 is high (AG, 24 kJ/mol= 5.7 kcallmol), correlating with its bigh melting temperature of 11I OT.

One of the physiological roles of 'OFn3 is as a subunit of fibronectin, a glycoprotein that exists in a soluble form in body fluids and in an insoluble 1- 0 0 -14-

("N

Sfonrm in the extracellular matrix (Dickinson et al., J. Mol. Biol. 236:1079, 00 1994). A fibronectin monomer of 220e250 liD contains 12 type I modules, two o type II modules, and 17 fibronectin type IH modules (Potts and Campbell, Curr.

Opin.Cell Biol. 6:648, 1994). Different type III modules are involved in the t> 5 binding of fibronectin to integrins, heparin, and chondroitin sulfate. '"Fn3 was Va 00 found to mediate cell adhesion through an integrin-binding Arg-Gly-Asp (RGD) motif on one of its exposed loops. Similar RGD motifs have been o shown to be involved in integrin binding by other proteins, such as fibrinogen,

C

N von Wellebrand factor, and vitronectin (Hynes et al., Cell 69:11, 1992). No other matrix- or cell-binding roles have been described for "°Fn3.

The observation that 'IFn3 has only slightly more adhesive activity than a short peptide containing RGD is consistent with the conclusion that the cell-binding activity of'Fn3 is localized in the RGD peptide rather than distributed throughout the 0 Fn3 structure (Baron et al., Biochemistry 31:2068, 1992). The fact that '°Fn3 without the RGD motif is unlikely to bind to other plasma proteins or extracellular matrix makes '°Fn3 a useful scaffold to replace antibodies. In addition, the presence of '°Fn3 in natural fibrinogen in the bloodstream suggests that '°Fn3 itself is unlikely to be immunogenic in the organism of origin.

In addition, we have determined that the 0 Fn3 framework possesses exposed loop sequences tolerant of randomization, facilitating the generation of diverse pools of antibody mimics. This determination was made by examining the flexibility of the "Fn3 sequence. In particular, the human '°Fn3 sequence was aligned with the sequences offibronectins from other sources as well as sequences of related proteins (Figure and the results of this alignment were mapped onto the three-dimensional structure of the human '"Fn3 domain (Figure This alignment revealed that the majority of conserved residues are found in the core of the beta sheet sandwich, whereas the highly variable 0 o residues are located along the edges of the beta sheets, including the N- and 0 0 C-termini, on the solvent-accessible faces of both beta sheets, and on three o solvent-accessible loops that serve as the hypervariable loops for affinity maturation of the antibody mimics. In view of these results, the randomization of these three loops are unlikely to have an adverse effect on the overall fold or 00 stability of the "'Fn3 framework itself.

For the human 0 Fn3 sequence, this analysis indicates that, at a ominimum, amino acids 1-9, 44-50, 61-54, 82-94 (edges of beta sheets); 19, 21, 30-46 (even), 79-65 (odd) (solvent-accessible faces of both beta sheets); 21-31, 51-56, 76-88 (CDR-like solvent-accessible loops); and 14-16 and 36-45 (other solvent-accessible loops and beta turns) may be randomized to evolve new or improved compound-binding proteins. In addition, as discussed above, alterations in the lengths of one or more solvent exposed loops may also be included in such directed evolution methods. Alternatively, changes in the I3sheet sequences may also be used to evolve new proteins. These mutations change the scaffold and thereby indirectly alter loop structure(s). If this approach is taken, mutations should not saturate the sequence, but rather few mutations should be introduced. Preferably, no more than 10 amino acid changes, and, more preferably, no more than 3 amino acid changes should be introduced to the P-sheet sequences by this approach.

Fibronectin Fusions The antibody mimics described herein may be fused to other protein domains. For example, these mimics may be integrated with the human immune response by fusing the constant region of an IgO with a 'Fn3 module, preferably through the C-terminus of "OFn3. The F. in such a "°Fn3-F, fusion molecule activates the complement component of the immune response and increases the therapeutic value of the antibody mimic. Similarly, a fusion

O

O -16t between '°Fn3 and a complement protein, such as Clq, may be used to target cells, and a fusion between 0 Fn3 and a toxin may be used to specifically o destroy cells that carry a particular antigen. In addition, "Fn3 in any form may be fused with albumin to increase its half-life in the bloodstream and its tissue S 5 penetration. Any of these fusions may be generated by standard techniques, for 00 example, by expression of the fusion protein from a recombinant fusion gene constructed using publically available gene sequences.

O0 Fibronectin Scaffold Multimers In addition to fibronectin monomers, any of the fibronectin constructs described herein may be generated as dimers or multimers of "'Fn3-based antibody mimics as a means to increase the valency and thus the avidity of antigen binding. Such multimers may be generated through covalent binding between individual m'Fn3 modules, for example, by imitating the natural 8 Fn3- 9 Fn3-'

O

Fn3 C-to-N-terminus binding or by imitating antibody dimers that are held together through their constant regions. A t"Fn3-Fc construct may be exploited to design dimers of the general scheme of "OFn3-Fc::Fc-'Fn3. The bonds engineered into the Fc::Fc interface may be covalent or non-covalent In addition, dimerizing or multimerizing partners other than Fc can be used in Fn3 hybrids to create such higher order structures.

In particular examples, covalently bonded multimers may be generated by constructing fusion genes that encode the multimer or, alternatively, by engineering codons for cysteine residues into monomer sequences and allowing disulfide bond formation to occur between the expression products. Non-covalently bonded multimers may also be generated by a variety of techniques. These include the introduction, into monomer sequences, of codons corresponding to positively and/or negatively charged -17- Sresidues and allowing interactions between these residues in the expression O products (and therefore between the monomers) to occur. This approach may 00 0 be simplified by taking advantage of charged residues naturally present in a monomer subunit, for example, the negatively charged residues of fibronectin.

Another means for generating non-covalently bonded antibody mimics is to 00 introduce, into the monomer gene (for example, at the amino- or carboxyc' termini), the coding sequences for proteins or protein domains known to o interact Such proteins or protein domains include coil-coil motifs, leucine zipper motifs, and any of the numerous protein subunits (or fragments thereof) known to direct formation of dimers or higher order multimers.

Fibronectin-.Like Molecules Although 'Fn3 represents a preferred scaffold for the generation of antibody mimics, other molecules may be substituted for '°Fn3 in the molecules described herein. These include, without limitation, human fibronectin modules 'Fn3-Fn3 and 'Fn3-"Fn3 as well as related Fn3 modules from non-human animals and prokaryotes. In addition, Fn3 modules from other proteins with sequence homology to 0 Fn3, such as tenascins and undulins, may also be used. Modules from different organisms and parent proteins may be most appropriate for different applications; for example, in designing an antibody mimic, it may be most desirable to generate that protein from a fibronectin or fibronectin-like molecule native to the organism for which a therapeutic or diagnostic molecule is intended.

Directed Evolution of Scaffold-Based Binding Proteins The antibody mimics described herein may be used in any technique for evolving new or improved binding proteins. In one particular example, the target of binding is immobilized on a solid support, such as a column resin or O -18- Smicrotiter plate well, and the target contacted with a library of candidate O scaffold-based binding proteins. Such a library may consist of '°Fn3 clones o constructed from the wild type '°Fn3 scaffold through randomization of the sequence and/or the length of the 'Fn3 CDR-like loops. If desired, this library S 5 may be an RNA-protein fusion library generated, for example, by the ,o techniques described in Szostak et al., U.S.S.N. 09/007,005 and 09/247,190; Szostak et al., W098/31700; and Roberts Szostak, Proc. Natl. Acad. Sci.

o USA (1997) vol. 94, p. 12297-12302. Alternatively, it may be a DNA-protein CN library (for example, as described in Lohse, DNA-Protein Fusions and Uses Thereof, US Patent No. 6,416,950. The fusion library is incubated with the immobilized target, the support is washed to remove non-specific binders, and the tightest binders are eluted under very stringent conditions and subjected to PCR to recover the sequence information or to a create a new library of binders which may be used to repeat the selection process, with or without further mutagenesis of the sequence. A number of rounds of selection may be performed until binders of sufficient affinity for the antigen are obtained.

In one particular example, the '°Fn3 scaffold may be used as the selection target. For example, ifa protein is required that binds a specific peptide sequence presented in a ten residue loop, a single 'OFn3 clone is constructed in which one of its loops has been set to the length of ten and to the desired sequence. The new clone is expressed in vive and purified, and then immobilized on a solid support. An RNA-protein fusion library based on an appropriate scaffold is then allowed to interact with the support, which is then washed, and desired molecules eluted and re-selected as described above.

Similarly, the 'tFn3 scaffold may be used to find natural proteins that interact with the peptide sequence displayed in a 'OFn3 loop. The '°Fn3 protein is immobilized as described above, and an RNA-protein fusion library is o -19- 0 screened for binders to the displayed loop. The binders are enriched through 0 0 In addition, in the above approaches, although RNA-protein libraries represent exemplary libraries for directed evolution, any type of scaffold-based Va 5 library may be used in the selection methods of the invention.

00 UMc o The antibody mimics described herein may be evolved to bind any antigen of interest. These proteins have thermodynamic properties superior to those of natural antibodies and can be evolved rapidly in slim. Accordingly, these antibody mimics may be employed in place of antibodies in all areas in which antibodies are used, including in the research, therapeutic, and diagnostic fields. In addition, because these scaffolds possess solubility and stability properties superior to antibodies, the antibody mimics described herein may also be used under conditions which would destroy or inactivate antibody molecules. Finally, because the scaffolds of the present invention may be evolved to bind virtually any compound, these molecules provide completely novel binding proteins which also find use in the research, diagnostic, and therapeutic areas.

Exemplary scaffold molecules described above were generated and tested, for example, in selection protocols, as follows.

[ibn xcosarution A complex library was constructed from three fragments, each of which contained one randomized area corresponding to a CDR-like loop. The fragments were named BC, DE, and PC], based on the names of the

O

o SCDR-H-like loops contained within them; in addition to 'OFn3 and a

O

o randomized sequence, each of the fragments contained stretches encoding an O N-terminal His, domain or a C-terminal FLAG peptide tag. At each junction between two fragments between the BC and DE fragments or between the 5 DE and FG fragments), each DNA fragment contained recognition sequences Sfor the Earl Type IIS restriction endonuclease. This restriction enzyme allowed 1 the splicing together of adjacent fragments while removing all foreign, 0 non-' 0 Fn3, sequences. It also allows for a recombination-like mixing of the three "'Fn3 fragments between cycles of mutagenesis and selection.

Each fragment was assembled from two overlapping oligonucleotides, which were first annealed, then extended to form the double-stranded DNA form of the fragment. The oligonucleotides that were used to construct and process the three fragments are listed below; the "Top" and "Bottom" species for each fragment are the oligonucleotides that contained the entire "'Fn3 encoding sequence. In these oligonucleotides designations, indicates A, T, C, or G; and indicates C or G.

HfnLbcTop (His): GG AAT TCC TAA TAC GAC TCA CTA TAG GGA CAA TTA CTA TTT ACA ATT ACA ATG CAT CAC CAT CAC CAT CAC GTT TCT GAT GTT CCG AGG GAC CTG GAA GTT GTT GCT GCG ACC CCC ACC AGC-3' (SEQ ID NO: 1) HfnLbcTop (an alternative N-terminus): GG AAT TCC TAA TAC GAC TCA CTA TAG GGA CAA TTA CTA TTT ACA ATT ACA ATG GTT TCT GAT GTT CCG AGG GAC CTG GAA GTT GTT GCT GCG ACC CCC ACC AGC-3' (SEQ ID NO: 2) o -21tHIFnLBCBct-flagS: 0 5'-AGC GGA TG CTT CrC GTC GTC OTC CTT GTA GTC OCT Cli? 00 o CCC TGT TTC TCC OTA ACT GAT CCT OTA ATA TCT (SNN)7 CGA OCT OAT CG TAG OCT OCT G00 GOT CCC AGC (SEQ ID NO: 3) 00 5 HF8BC3'-flag8: OGA TG CIT GTC GTC GTC GTG CTT OTA GTC OCT CT o CCC TOT rrC TCG GIA AGT OAT CC-3 (SEQ ID NO: 4) HFnLDETop: 00 NAT TCC TAA TAC GAG TCA CTA TAG OGA CAA i'TA CTA TTn AGA AU? ACA ATO CAT GAG CAT GAG CAT GAG CTC TITC ACA OGA OGA. AAT AGC GCT GTC C-3' (SEQ ID NO: FnLDEBot-flag8: OGA TG Cii? GTC GTC CTC GTC Cii? OTA OTCGOCT CiT GT ATA ATG AAC TGC AGO rrr AAG 0CC OCT GAT OCT AGC TOT (SNN)4 AGO, GAG AGT GAA GTG CTG GAG AGO GCT AU? TCC TCG TOT (SEQ ID NO: 6) HFcDE3'-flagS: TG CTT GTC OTC GTG GTC CTI? OTA GTCGOCT CIT COT ATA ATG AG TGC AOG iii? AAG 0-3' (SEQ ID NO: 7) HFmLFGTop: 00 NAT TGC TAA TAG GAG TCA CTA TAG OGA CAA TTA CTA TTT ACA AlT ACA ATO CAT GAG CAT GAG CAT GAG CTG TTC TAT ACC ATC ACT GTG TAT OCT OTC-3' (SEQ LD NO: 8) t HFnLFGBot-flag8: 005'-AGC (3CA TOC CIT GTC (3TC GTC GTC CT]? OTA GTC TOT TCGi o OTA A2FT AAT OGA AAT TOG (SNN) 10 ACT GAC AGC ATA CAC AGT OAT GOT ATA (SEQ ID NO: 9) 00 5 HFnFG3'-flag8: GOA TOC CiT GTC GTC GTC GTC CiT OTA GTC TGT TCG o (HAA MT AAT GGA AAT TOG (SEQ ID) NO: T7Tmv (introduces T7 promoter and TMV untranslated region needed for In vitro translation): OCO TAA TAC GAC TCA CTA TAG OGA CAA TITA CTA liT ACA AlT ACA-3' (SEQ ID NO: 11) ASAflag8: OGA TG CTT GTC GTC GTC GTC CT]? OTA GTC-3' (SEQ ID NO: 12) Uispl-s (spint oligomucleotide used to ligate, nRNA to the puromycin-contaiming linker, described by Roberts et al, 1997, supra): 5'-I-ITTJl''TFNAGCGGATGC-3' (SEQ ID NO: 13) AIS-2PEG (DNA-puroinycln linker): I 8(PEG)2CCPur (SEQ ID NO: 14) The pairs of oligonucleotides (500 pmol of each) were annealed in 100 g.L of 10 mM Tris 7.5, 50 mM NaCI for 10 minutes at 85 0 C, followed by a slow (0.5-1 hour) cooling to room temnperature. The annealed fragments with

O

O -23- Ssingle-stranded overhangs were then extended using 100 U Klenow (New O England Biolabs, Beverly, MA) for each 100 pL aliquot of annealed oligos, and 00 oo O the buffer made of 838.5 pl H20, 9 Il I M Tris 7.5, 5 1l 1M MgCIl, 20 1 A mM dNTPs, and 7.5 ul 1M DTT. The extension reactions proceeded for I hour at 00 Next, each of the double-stranded fragments was transformed into a ¢c RNA-protein fusion (PROfusion

T

using the technique developed by Szostak Set al., US Patent No. 6,258,558 and US Patent No. 6,261,804; Szostak et al.

0 W098/31700; and Roberts Szostak, Proc. Natl. Acad. Sci. USA (1997) vol.

94, p. 12297-12302. Briefly, the fragments were transcribed using an Ambion in itro transcription kit, MEGAshortscript (Ambion, Austin, TX), and the resulting mRNA was gel-purified and ligated to a DNA-puromycin linker using DNA ligase. The mRNA-DNA-puromycin molecule was then translated using the Ambion rabbit reticulocyte lysate-based translation kit. The resulting mRNA-DNA-puromycin-protein PROfusion T was purified using Oligo(dT) cellulose, and a complementary DNA strand was synthesized using reverse transcriptase and the RT primers described above (Unisplint-S or flagASA), following the manufacturer's instructions.

The PROfusionT obtained for each fragment was next purified on the resin appropriate to its peptide purification tag, on Ni-NTA agarose for the His,-tag and M2 agarose for the FLAG-tag, following the procedure recommended by the manufacturer. The DNA component of the tag-binding PROfusions was amplified by PCR using Pharmacia Ready-to-Go PCR Beads, 10 pmol of 5' and 3' PCR primers, and the following PCR program (Pharmacia, Piscataway, NJ): Step 1: 95*C for 3 minutes; Step 2: 95°C for seconds, 58/62*C for 30 seconds, 72"C for 1 minute, 20/25/30 cycles, as required; Step 3: 72C for 5 minutes; Step 4: 4°C until end.

The resulting DNA was cleaved by 5 U Earl (New England Biolabs) 0 0 -24- 0 perl ug DNA; the reaction took place in T4 DNA Ligase Buffer (New England Biolabs) at 37°C, for I hour, and was followed by an incubation at 70'C for 0 minutes to inactivate Ear I. Equal amounts of the BC, DE, and FG fragments were combined and ligated to form a full-length "OFn3 gene with randomized 5 loops. The ligation required 10 U of fresh Earl (New England Biolabs) and U of T4 DNA Ligase (Promega, Madison, WI), and took 1 hour at 37°C.

Three different libraries were made in the manner described above.

o Each contained the form of the FG loop with 10 randomized residues. The BC and the DE loops of the first library bore the wild type "'Fn3 sequence; a BC loop with 7 randomized residues and a wild type DE loop made up the second library; and a BC loop with 7 randomized residues and a DE loop with 4 randomized residues made up the third library. The complexity of the FG loop in each of these three libraries was 10"; the further two randomized loops provided the potential for a complexity too large to be sampled in a laboratory.

The three libraries constructed were combined into one master library in order to simplify the selection process; target binding itself was expected to select the most suitable library for a particular challenge.

PROfusions T were obtained from the master library following the general procedure described in Szostak et al., U.S.S.N. 09/007,005 and 09/247,190; Szostak et al., W098/31700; and Roberts Szostak, Proc. Nati. Acad. Sci.

USA (1997) vol. 94, p. 12297-12302 (Figure 8).

Fusion Selecins The master library in the PROfusionT M form was subjected to selection for binding to TNF-a. Two protocols were employed: one in which the target was immobilized on an agarose column and one in which the target was immobilized on a BIACORE chip. First, an extensive optimization of conditions to minimize background binders to the agarose column yielded the 0 O t favorable buffer conditions of 50 mM HEPES pH 7.4, 0.02% Triton, 100 gg/ml Sheared Salmon Sperm DNA. In this buffer, the non-specific binding of the O 'oFn3 RNA fusion to TNF-a Sepharose was The non-specific binding background of the '°Fn3 RNA-DNA to TNF-a Sepharose was found to be 5 0.1%.

0 During each round of selection on TNF-a Sepharose, the Profusion

T

Slibrary was first preincubated for an hour with underivatized Sepharose to o remove any remaining non-specific binders; the flow-through from this preclearing was incubated for another hour with TNF-a Sepharose. The TNF-a Sepharose was washed for 3-30 minutes.

After each selection, the PROfusionT DNA that had been eluted from the solid support with 0.3 M NaOH or 0.1M KOH was amplified by PCR; a DNA band of the expected size persisted through multiple rounds of selection (Figure similar results were observed in the two alternative selection protocols, and only the data from the agarose column selection is shown in Figure 9.

In the first seven rounds, the binding of library PROfusions" T to the target remained low; in contrast, when free protein was translated from DNA pools at different stages of the selection, the proportion of the column binding species increased significantly between rounds (Figure 10). Similar selections may be carried out with any other binding species target (for example, IL-1 and L-13).

Animal Studies Wild-type 'OFn3 contains an integrin-binding tripepetide motif, Arginine 78 Glycine 79 Aspartate 80 (the "RGD motif) at the tip of the FG loop. In order to avoid integrin binding and a potential inflammatory response based on this tripeptide in ivo, a mutant form of 'oFn3 was generated that

O

O -26o contained an inert sequence, Serine 78 Glycine 79 Glutamate 80 (the "SGE O mutant"), a sequence which is found in the closely related, wild-type "Fn3 00 O domain. This SGE mutant was expressed as an N-terminally His 6 -tagged, free protein in E coli, and purified to homogeneity on a metal chelate column In 5 followed by a size exclusion column.

00 In particular, the DNA sequence encoding His 6 0 Fn3(SGE) was C cloned into the pET9a expression vector and transformed into BL21 DE3 o pLysS cells. The culture was then grown in LB broth containing 50 g/mL kanamycin at 37°C, with shaking, to Am= 1.0, and was then induced with 0.4 mM IPTG. The induced culture was further incubated, under the same conditions, overnight (14-18 hours); the bacteria were recovered by standard, low speed centrifugation. The cell pellet was resuspended in 1/50 of the original culture volume of lysis buffer (50 mM Tris 8.0, 0.5 M NaCI, glycerol, 0.05% Triton X-100, and 1 mM PMSF), and the cells were lysed by passing the resulting paste through a Microfluidics Corporation Microfluidizer M110-EH, three times. The lysate was clarified by centrifugation, and the supernatant was filtered through a 0.45 pm filter followed by filtration through a 0.2 pm filter. 100 mL of the clarified lysate was loaded onto a 5 mL Talon cobalt column (Clontech, Palo Alto, CA), washed by 70 mL of lysis buffer, and eluted with a linear gradient of 0-30 mM imidazole in lysis buffer. The flow rate through the column through all the steps was I mL/min. The eluted protein was concentrated 10-fold by dialysis (MW cutoff 3,500) against 15,000-20,000 PEG. The resulting sample was dialysed into buffer 1 (lysis buffer without the glycerol), then loaded, 5 mL at a time, onto a 16 x 60 mm Sephacryl 100 size exclusion column equilibrated in buffer 1. The column was run at 0.8 mL/min, in buffer 1; all fractions that contained a protein of the expected MW were pooled, concentrated 10X as described above, then dialyzed into PBS. Toxikon (MA) was engaged to perform endotoxin screens 0 o -27t and animal studies on the resulting sample.

O In these animal studies, the endotoxin levels in the samples examined 00 0 to date have been below the detection level of the assay. In a preliminary toxicology study, this protein was injected into two mice at the estimated 100X therapeutic dose of 2.6 mg/mouse. The animals survived the two weeks of the

\O

00 study with no apparent ill effects. These results suggest that '°Fn3 may be Sincorporated safely into an IV drug.

0 Alternative Constructs for In Vivo Use To extend the half life of the 8 kD 'Fn3 domain, a larger molecule has also been constructed that mimics natural antibodies. This '"Fn3-F, molecule contains the -CHi-CH 2

-CH

3 (Figure 11) or -CH 2

-CH

3 domains of the IgG constant region of the host; in these constructs, the 'OFn3 domain is grafted onto the N-terminus in place of the IgG V, domain (Figures 11 and 12). Such antibody-like constructs are expected to improve the pharmacokinetics of the protein as well as its ability to harness the natural immune response.

In order to construct the murine form of the '°Fn3-CH,-CH 2

-CH

3 clone, the -CH,-CH 2 -CH, region was first amplified from a mouse liver spleen cDNA library (Clontech), then ligated into the pET25b vector. The primers used in the cloning were 5' Fe Nest and 3' 5 Fc Nest, and the primers used to graft the appropriate restriction sites onto the ends of the recovered insert were Fc HIII and 3' Fc Nhe: Fe Nest 5'GCG GCA GGG TTT GCT TAC TGG GGC CAA GGG 3' (SEQ ID NO: 3' Fc Nest 5'GGG AGG GGT GGA GGT AGG TCA CAG TCC 3' (SEQ ID NO: 16); 3' Fc Nhe 5' TTT GCT AGC TTT ACC AGG AGA GTG GGA GGC 3' (SEQ -28- ID NO: 17); and o 5' Fc HIII 5' AAA AAG CTT GCC AAA ACG ACA CCC CCA TCT GTC 3' (N (SEQ ID NO: 18).

0 0 o Further PCR is used to remove the CH, region from this clone and create the Fc part of the shorter, "'Fn3-CH 2 -CH, clone. The sequence encoding '°Fn3 is spliced onto the 5' end of each clone; either the wild type '°Fn3 cloned from the same mouse spleen cDNA library or a modified '°Fn3 obtained by 0. mutagenesis or randomization of the molecules can be used. The 4 oligonucleotides used in the cloning of murine wild-type ToFn3 were: 0 Mo CATATGGTTTCTGATATTCCGAGAGATCTGGAG 3' (SEQ ID NO: 19); (for an alternative N-terminus with the His purification tag): CAT ATG CAT CAC CAT CAC CAT CAC GTT TCT GAT ATT CCG AGA G 3' (SEQ ID NO: 20); and Mo3PCR-EcoRI: GAATTCCTATGTTTTATAATTGATGGAAAC3' (SEQ ID NO: 21).

The human equivalents of the clones are constructed using the same strategy with human oligonucleotide sequences.

Other embodiments are within the claims.

All publications, patents, and patent applications mentioned herein are hereby incorporated by reference.

Throughout this specification and the claims which follow, unless the context requires otherwise, the word "comprise", and variations such as "comprises" and "comprising", will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps.

The reference to any prior art in this specification is not, and should not be taken as, an acknowledgement or any form of suggestion that the prior art forms part of the common general knowledge in Australia.

Claims

1. A protein comprising: t\ 0(a) a first fibronectin type III (Fn3) domain, wherein said first Fn3 domain has an amino acid sequence that has been altered in one or more loops relative to the amino acid sequence of a Vt') r- first human Fn3 domain such that the first Fn3 domain binds to a first compound that is not 00 bound by the first human Fn3 domain; and a second Fn3 domain, wherein said second Fn3 domain has an amino acid sequence that CI has been altered in one or more loops relative to the amino acid sequence of a second human Fn3 domain such that the second Fn3 domain binds to a second compound that is not bound by the second human Fn3 domain.

2. The protein of claim 1, wherein said protein further comprises one or more additional Fn3 domains, and wherein each of said additional Fn3 domains has an amino acid sequence that has been altered in one or more loops relative to the amino acid sequence of a human Fn3 domain such that each said additional Fn3 domain binds to a compound that is not bound by the corresponding human Fn3 domain.

3. The protein of claim 1 or 2, wherein said Fn3 domains are the tenth Fn3 module of human fibronectin (10Fn3), wherein the 10Fn3 domain comprises an AB loop, a BC loop, a DE loop, and an FG loop and has at least one loop selected from the BC loop, the DE loop, and the FG loop with an altered amino acid sequence relative to the sequence of the corresponding loop of the human 10Fn3 domain.

4. The protein of claim 2, wherein said protein comprises: a first Fn3 domain wherein said first Fn3 domain is an eighth Fn3 module of human fibronectin (8Fn3) and has an amino acid sequence that has been altered in one or more loops relative to the amino acid sequence of human (8Fn3) such that the first Fn3 domain binds to a first compound that is not bound by human 8Fn3; 00 O a second Fn3 domain wherein said first Fn3 domain is a ninth Fn3 module of human N1 fibronectin (9Fn3) and has an amino acid sequence that has been altered in one or more loops relative to the amino acid sequence of human (9Fn3) such that the second Fn3 domain binds to a second compound that is not bound by human 9Fn3; and a third Fn3 domain wherein said third Fn3 domain is a tenth Fn3 module of human fibronectin (10Fn3) and has an amino acid sequence that has been altered in one or more 00 loops relative to the amino acid sequence of human (10Fn3) such that the third Fn3 domain CI binds to a first compound that is not bound by human 10Fn3. C N

5. The protein of any one of the preceding claims further comprising a second polypeptide, wherein the second polypeptide is selected from the group consisting of an Fc region of an antibody, a complement protein, an albumin protein, and a toxin protein.

6. The protein of any one of the preceding claims, wherein the protein is substantially free of endotoxin.

7. A pharmaceutical composition substantially free of endotoxin comprising a therapeutic protein, wherein the protein comprises a fibronectin type III (Fn3) domain, wherein the Fn3 domain: has at least one loop with an altered amino acid sequence relative to the sequence of the corresponding loop of a human Fn3 domain and binds to a target compound that is not bound by the corresponding human Fn3 domain.

8. The pharmaceutical composition of claim 7, wherein the Fn3 domain is a 10Fn3 domain comprising an AB loop, a BC loop, a DE loop, and an FG loop; and wherein the 10Fn3 domain has at least one loop selected from the BC loop, the DE loop, and the FG loop with an altered amino acid sequence relative to the sequence of the corresponding loop of the human 10Fn3 domain.

9. The pharmaceutical composition of claim 7 or 8, wherein the Fn3 domain has at least two loops with an altered amino acid sequence relative to the sequence of the corresponding loop of a human Fn3 domain. 00 0

10. The pharmaceutical composition of claim 9, wherein the Fn3 domain has at least three N loops with an altered amino acid sequence relative to the sequence of the corresponding loop of a human Fn3 domain.

11. The pharmaceutical composition of any one of claims 7-10, wherein the therapeutic protein further comprises an Fc region of an antibody. 00

12. The pharmaceutical composition of claim 11, wherein the Fc region is an IgG Sconstant region.

13. The pharmaceutical composition of claim 8, wherein the integrin binding motif, arginine-glycine-aspartic acid (RGD), of the lOFn3 domain is replaced by an amino acid sequence as follows: basic amino acid-neutral amino acid-acidic amino acid.

14. A pharmaceutical composition comprising a protein of any one of claims 1-6 and a physiologically acceptable carrier, wherein the pharmaceutical composition is substantially free of endotoxin.

A molecule comprising i) a protein comprising a fibronectin type III (Fn3) domain, wherein the Fn3 domain: has at least one loop with an altered amino acid sequence relative to the sequence of the corresponding loop of a human Fn3 domain; and binds to a target compound that is not bound by the corresponding human Fn3 domain; and ii) a ribonucleic acid; wherein the protein is bonded through a DNA-puromycin linker to the ribonucleic acid, and wherein the protein is encoded by said ribonucleic acid.

16. A molecule of claim 15, wherein the Fn3 domain is a tenth type Fn3 domain (10Fn3), and wherein the 10Fn3 domain comprises an AB loop, a BC loop, a DE loop, and an FG loop and has at least one loop selected from the BC loop, the DE loop, and the FG loop with an altered amino acid sequence relative to the sequence of the corresponding loop of the human 1 OFn3 domain. 00

17. A molecule of claim 16, wherein the 10Fn3 domain has at least two loops with an 0 altered amino acid sequence relative to the sequence of the corresponding loop of a human 1OFn3 domain.

18. A molecule of claim 17, wherein the 10Fn3 domain has at least three loops with an altered amino acid sequence relative to the sequence of the corresponding loop of a human t 10OFn3 domain. 00

19. A molecule of any one of claims 16-18, wherein the integrin binding motif, arginine- glycine-aspartic acid (RGD), of the 10 OFn3 domain is replaced by an amino acid sequence as follows: basic amino acid-neutral amino acid-acidic amino acid.

The molecule of any one of claims 15-19 selected by the method comprising the steps of a) producing a population of candidate RNA molecules, each comprising a candidate fibronectin type III (Fn3) domain sequence which differs from human Fn3 domain coding sequence, said RNA molecules each comprising a translation initiation sequence and a start codon operably linked to said candidate Fn3 domain coding sequence and each being operably linked to a DNA-puromycin linker at the 3' end; b) in vitro translating said candidate Fn3 domain coding sequences to produce a population of candidate RNA-Fn3 fusions; c) contacting said population of candidate RNA-Fn3 fusions with the target compound; and d) selecting an RNA-Fn3 fusion, the protein portion of which has a binding affinity for said target compound that is altered relative to the binding affinity of said human Fn3 for said target molecule.

21. The molecule of claim 20, wherein the selection method further comprises the steps of e) altering the Fn3 domain of the RNA-Fn3 fusions isolated from step f) contacting said population of candidate RNA-Fn3 fusions from step e) with the target compound; and g) selecting an RNA-Fn3 fusion, the protein portion of which has a binding affinity for said target compound that is altered relative to the binding affinity of said Fn3 from step d) for said target molecule. 00

22. The protein according to claim 1, or the pharmaceutical composition according to claim 7, or molecule of claim 15, substantially as herein described with reference to the 4. Examples and/or Drawings. 00 00 i", i",